Beam model by Reflection is now available on Amazon SageMaker JumpStart and AWS Trainium

Reflection’s Beam model, an open-weight frontier AI system, has been added to Amazon SageMaker JumpStart, the model hub within Amazon SageMaker AI. The integration allows customers to deploy Beam alongside existing models while maintaining ownership of their data and infrastructure. SageMaker’s built-in controls, audit trails, and monitoring support scalable production use without requiring changes to existing workflows.

Beam was specifically ported and optimized for AWS Trainium3, providing a high-performance, cost-efficient deployment option from launch. The optimization leverages Trainium3’s hardware to deliver efficient inference and training while preserving the model’s open-weight flexibility. Customers can now run Beam in their own AWS accounts, ensuring data privacy and compliance with existing security measures.

Engineering teams can use Beam to automate code generation, review, and issue resolution, while businesses can apply it to multi-step workflows and advanced reasoning tasks. The model’s Apache 2.0 license permits inspection, modification, and customization, aligning with enterprise needs for transparency and adaptability. Deployment is supported in US East (N. Virginia) and US East (Ohio) regions via SageMaker JumpStart.

To access Beam, users can search for the model in SageMaker JumpStart within Amazon SageMaker Studio or deploy it directly on SageMaker AI. Additional resources include the AWS/Beam launch blog, SageMaker AI documentation, and SageMaker JumpStart documentation for implementation guidance. The release underscores a shift toward self-hosted, controllable AI systems in enterprise environments.